Preliminary Recommendations

 

Background and objectives

The specification of subcategorisation structure is perhaps the most urgent and complex type of basic linguistic information which a lexicon suitable for NLP applications must provide. Urgency arises from the need to encode information concerning the local context of words in order to constrain the analysis and generation of natural language. Complexity results from the multidimensional aspect of subcategorisation structure. Because subcategorisation involves reference to diverse levels of grammatical description, the achievement of efficient standards in this area presupposes a minimal normalisation of

• Morphological, syntactic and semantic aspects of word specification; and
• The interface between these levels of grammatical description.

The standardisation of subcategorisation is therefore a crucial task in the provision of a protocol for

• Comparing and exchanging existing lexical resources; and
• Building new monolingual and multilingual lexica which are compatible with or easily adaptable to different grammatical platforms.

Work on subcategorisation has been carried out from a variety of perspectives by linguists, lexicographers and developers of NLP applications for a variety of languages. Consequently, the formulation of standards in this area necessitates a comparison of linguistic theories, practical NLP lexicons, dictionaries and bracketted corpora as well as language-specific aspects of subcategorisation structure. The main aim of this document is to then assess how and to what extent such a comparison can be carried out in order to achieve an inter-theoretical specification which captures grammatical equivalences among system- or theory-specific coding schemes for the main European languages.

In carrying out this work, our goals are to:

• Define a scheme which standardises approaches to subcategorisation from --
  • Linguistic theories (GB, HPSG, LFG, Unification Categorial Grammar, Dependency Grammar)
  • Practical NLP lexicons (ACQUILEX, COMLEX, EUROTRA, GENELEX, ILC, PLNLP)
  • Bracketted corpora (UPENN, TOSCA, SUSANNE and HELSINKI treebanks)
• Prove the validity of our standardisation scheme for a variety of European languages.